Serena Foley, a chemical and biopharmaceutical student at Cork Institute of Technology, describes her final-year research project that identified factors contributing to downtime during changeover operations at an API site and recommended preventative measures.

The following article represents areas identified to improve changeover operations which can help mitigate the downtime experienced when manufacturing APIs on a batch basis. The data collected to accomplish such a project was extracted from Pfizer Ireland Pharmaceuticals, API Plant, Ringaskiddy Cork.

Based on a single production plant within their API site in Ringaskiddy, the planning and execution of ‘changeovers’ were examined based on their effectiveness, efficiency and overall performance.

Cleaning and preparation of equipment vessels

Changeovers involve the cleaning and preparation of equipment vessels once made available after the discharge and offload of the last batch of an API/intermediate and before the charge of the subsequent new API/intermediate campaign.

The main objective of the project was to reduce downtime during changeovers whilst achieving the required quality specifications.  

Table 1 – ‘Bottleneck’ delays identified between 2016 to 2019, extracted from Annual Digital Delay Trackers

Any downtime experienced on plant is tracked via bottleneck delay trackers; documenting length of time lost, identifying responsible personnel, root cause for delay and detailing the required actions which must ensue.

Such annual delay trackers were analysed from 2016 to 2019 where the overall downtime experienced was broken down into being either production process or changeover related.

Table 1 above highlights that in 2019 alone, a staggering 40.5% of total downtime experienced was due to unexpected delays within changeover operations. The root causes for these delays are shown in Figure 1 below:

Figure 1, above – Pie chart illustrating root causes for changeover downtime in 2019.

Figure 1 highlights how the majority of the delays within changeover were a result of repeating sampling and testing when initial runs failed the required cleaning quality specifications.

Rinse checks are a testing requirement used where the validity of cleaning is established and compared to an appropriate GMP protocol. Every vessel in an equipment train has a control point where a sample of solvent or rinse water is taken and tested for product residue, unknowns and solution clarity. If a sample along with its subsequent investigatory samples fail any of the specifications, an investigation is opened.

New samples and re-tests are conducted. Each re-test can cause delays of up to six hours. The accumulation of these unnecessary six-hour delays resulted in the most downtime experienced within changeover operations.

If the initial sampling and testing can pass the quality validation specifications ‘right first time’ (RFT) i.e. all control points within the equipment train pass specifications (100%), the need to ever perform re-sampling and re-testing can be eliminated. This results in a reduction of overall production downtime on plant.

Data preparation and statistical analysis

The aim is to achieve a high RFT metric where instances of failing specifications and nonconformities are diminished using a six-sigma approach. 

Regarding rinse checks, data from a sample group of 68 changeovers performed between 2016 and 2019 were populated based on the cleaning methodology used in the respective cleaning documentation. Such data extracted from the documentation included the following:

• Product name/number
• Cleaning solvent type,
• Total agitation time experienced for entire equipment train,
• Length of total recirculation time experienced for entire equipment train – includes cleaning cycles, rinse check cycles and final de-ionised water flushes,
• Length of refluxing time experienced for entire equipment train,
• The Right First Time (RFT) metric i.e. the number of control points passing the rinse check first time out of the total number of control points within the equipment train,
• The number of vessels per changeover and whether deemed as small (≤ 7 vessels), medium (8 ≤ ‘x’ ≤10 vessels) and large (≥ 11 vessels).

Three factors which varied across all cleaning approaches were; total agitation time per equipment train, length of recirculation time per equipment train and the length of reflux time per equipment train.

An investigation to determine whether any/all the factors mentioned above had a significant effect on the RFT metric for changeovers was performed. A sample of the data populated for such a study is shown in Table 2 below:Table 2 – Sample of data used for statistical analysis.

The hours calculated per investigated factor were distinguished into high (orange) and low (blue) categories based on the corresponding means within each year, as depicted within Table 2 and by the numbers ‘2’ and ‘1’ in Figures 2 and 3 below.

A Three-Way ANOVA study using the extracted data from the 68 changeovers was performed using the statistical software package Minitab19. Results are shown in Figure 2 and Figure 3:

Figure 2, above – Extent of the effect of the investigated factors on the mean of the RFT pass rate of rinse check

Figure 3, above – Interaction plots of all investigated factors vs. the mean of the RFT pass rate of rinse checks

Figure 2 shows that the recycling time has the largest effect on the RFT pass rate metric. Figure 3 demonstrates the possible interaction plots across all three investigated factors, highlighting that all interactions have an effect on the RFT metric as no pair of lines are parallel.

This statistical analysis resulted in the Ρ-value for the factor of recycle time is 0.05, which is equal to the significance level of 5% used within the study. This suggests that a significant effect does exist between the length of recirculation time and the RFT metric.

However, an outlier test was performed to analyse if any outliers existed within the 68 changeovers. Figure 4 highlights that two outliers were presented. The low RFT results within the 40% - 50% range were due to special cases.

This included an undetected solvent line which remained connected to a vessel where the control valve passed periodically, inadequate recirculation of contents for the specified time and gross decontamination on a vessel sampler was incomplete.

For these reasons, the two outlier RFT results were removed from the data set and the Three-Way ANOVA was performed again. The resulting Ρ-value for the recycle factor increased from 0.05 to 0.302.

This jump proves how only 2 data points within the original set of 68 can shift the findings of the study from in favour of the alternative hypothesis to in favour of the null hypothesis i.e. establishing if the recycling times have a substantial effect on the RFT metric and thus help minimise downtime experienced.

Figure 4, above – Outlier test on the RFT data for the 68 changeovers

Conclusions and recommendations

• Agitation time did not significantly affect RFT metric and changeover time.
• Reflux duration did not significantly affect RFT metric and changeover time.
• Recirculation time did does not significantly affect RFT metric and changeover time.
• Explore if the time for the above processes/factors could be reduced without compromising quality.
• Obtain more data to increase sample size and establish more conclusive findings.
• Initiate a trial run of increasing recirculation time by 50%. Monitor whether the RFT metric improves during future changeovers. This recommendation was suggested on the basis that the recirculation time was still the lowest calculated Ρ-value out of all factors and their interactions, despite it being greater than the significance level of 5% within the second ANOVA study.

Acknowledgements

I would like to sincerely thank Robert Kennedy as my industrial project supervisor within Pfizer Ringaskiddy and Dr. Cilian Ó Súilleabháin as my academic supervisor. Their guidance and feedback were appreciated throughout the course of the project.